Pain Assessment in Neonatal Clinical Practice via Facial Expression Analysis and Deep Learning

Since newborns are unable to verbally communicate the experience of pain, accurate pain assessment using validated tools is crucial to determine the most effective pain management strategies. Traditional pain assessment relies on the use of pain scales that consider behavioural, physiological, and contextual indicators, but is highly depending on the subjective experience of healthcare professionals. Therefore, automated pain assessment systems are desirable in clinical practice for a more efficient and objective pain evaluation. This work proposes a deep learning framework for newborn pain assessment based on facial expression analysis, suitable for the real-world environment of the Neonatology Department. It exploits a pre-trained Convolutional Neural Network that is fine-tuned on image data recorded during neonatal heel prick. The trained model achieves an average accuracy of 87.4% and F1-score of 75.4% using a stratified 5-fold cross-validation to classify newborn images into pain or nonpain classes. The performance is further improved by including images from iCOPE dataset in the training process. Moreover, the application of a visual explanation technique highlights that the model predictions are based on the facial regions most closely associated to the pain experience. Our results show that automated pain detection from facial expressions is feasible in a real-world setup. By introducing an explainable artificial intelligence approach, we also contribute to the improvement of model transparency and trust for healthcare professionals. This paves the way for the development of an automated system that integrates, standardises, and improves human pain evaluation.